YOLOv5 with Channel Attention and Feature Fusion for Wheel Surface Defect Detection

Wheels play an important role in automobiles. However, wheel surface defects may have an impact on the safety of the vehicle. Therefore, there is a need to identify the wheel defects. Unlike other common defects, there is a wide variety of wheel surface defects, and the wheel surface contains curved surfaces, which present a challenge for detection. In response to these problems, this paper proposes a wheel surface defect detection algorithm based on an improved YOLOv5. The improved algorithm embeds the efficient channel attention mechanism (ECA) and the adaptive spatial feature fusion (ASFF) detection head into the network and uses the structure of GSConv+SlimNeck. The improvement enables the algorithm to select the best convolutional kernel during feature extraction and fully fuse the features of different scales, enhancing the detection effect of small objects. The experimental results show that the improved YOLOv5 algorithm achieves higher mAP and detection accuracy while maintaining a high FPS, which meets the requirements for real-time detection.